Interpreting svyset for complex sampling design

[Rachel Rose]

Greetings,

I'm am having some difficulties determining how to declare my survey design with the svyset command. It is less a question about syntax and more a question regarding survey interpretation. I'm picking this up from someone else and guidance would be greatly appreciated. I have a survey with 600 samples for an evaluation that was sampled on these terms:
(1) At the first level 6 districts were selected for project implementation.
(2) Within each district a number of VDCs (governing unit) were selected (ranges from 2-6 VDCs per district) for a total sample of 20 VDCs. Half of the VDCs were assigned treatment status and half were assigned control status.
(3) Within each VDC a number of wards (smaller governing unit) were selected (varied from 2-3 wards per VDC) for a total of 57 wards.
(4) Finally a sample of households were selected within each ward for survey and were stratified based on light or intense treatment option (for those located within treatment VDCs) for a total of 600 households sampled.

I've calculated fpc values for each stage and tried to use these stages as described in svyset but I believe I'm doing something wrong because when I try to use svy estimation commands I get a note of "Missing standard errors because of stratum with single sampling unit." However, the whole reason for using svyset is get accurate standard errors given a complex sampling design. My interpretation on how to enter this information (and thus my code) using svyset is
svyset district_name, fpc(fpc_district) vce(linearized) singleunit(missing) || vdc_name, strata(treat_cont) fpc(fpc_vdc) || ward_number, fpc(fpc_ward) || household_number, strata(hh_type) fpc(fpc_hh)

Any guidance on interpretation of this sample design and thus svy set would be greatly appreciated.

Thank you!
Rachel

[Carole J. Wilson]

Rachel, I haven't read closely enough to comment on your actual svyset command, but the error you receive is because at least one of your strata has a single sampling unit. This may occur because of the design, the way you've specified the svyset, or because of list-wise deletion (or selection) in your estimation commands. In order to address the issue, you need to determine which strata are problematic.

To identify the offending strata:

1. run your estimation command (svy: regress...)
2. list the strata that contain only one unit: svydescribe if e(sample), single

Possible solutions:

• If the problem is due to the way you've specified your design, then fix it
• If it is due to missing values: inspect your variables to see if there is a problem in your coding or if you have other variables with fewer missing values that can serve as a proxy; look into ways to deal with missing values (multiple imputation)
• Drop the problematic strata (probably don't want to do that)
• If the strata are geographic, then it may be reasonable to combine the strata with a neighbor

[Rachel Rose]

Many thanks Carole. I think the issue is certainly without I've specified my design. I'm working to correct my mistake though am still have some issues with my interpretation of the survey design and how I designate that within Stata. Your command for svydescribe if e(sample), single was helpful!

[Steve Samuels]

Welcome to Statalist, Rachel! Please read FAQ 14 and always put code, results, and data listings between CODE delimiters.

Some ambiguity here: were the six districts selected by sampling from a larger population of districts? If not, how were they chosen? Until you clarify, I can't write down a correct svyset.

I also notice that pweight option is missing from your svyset statement. What is the name of the weight variable?